水稻生长模型参数的敏感性及其对产量遥感估测的不确定性

Sensitivity of rice growth model parameters and their uncertainties in yield estimation using remote sensing date

采用全局敏感性分析方法探讨水稻生长模型与遥感数据耦合估产的不确定性问题。以实测数据为基准数据,当ORYZA2000模型的17个输入变量模拟导入可能误差时,模型的地上生物量(WAGT)、叶面积指数(LAI)、籽粒质量(WSO)和叶片氮含量(NFLV)等模拟输出结果显示较大的不确定性,LAI最大变幅超过20%,最终籽粒产量最大变幅超过10%。引起模型输出结果不确定性的输入变量中,水稻播种期的影响最大;模型的驱动变量温度和日照时数的误差对成熟期的产量影响较大;水稻干物质地上叶片质量分数(FLVTB)对所有关于叶片和籽粒生物量的输出结果都有较大的影响,因此,当利用生长模型和遥感数据进行耦合时,水稻播种期、模型的驱动变量如温度、日照时数、FLVTB等数据精度对估产结果有很大影响。比较LAI、NFLV单独或LAI+NFLV同时耦合ORYZA2000模拟中,LAI+NFLV的敏感性指数最高,其次是LAI,NFLV最差。模型耦合估测WSO、WAGT时,水稻移栽后70～80d左右的遥感影像数据是最重要的,必须获得,此期前后20～30d的2次数据也比较重要,而幼苗期和成熟期的遥感数据耦合生长模型对产量和生物量的估测意义不大。

Uncertainty in output of ORYZA2000 （a rice growth model） and sensitivities of model inputs were analyzed through global sensitivity analysis. The actual measured data was as the reference values, used for describing uncertainty in the model inputs. Results revealed high uncertainties in model output such as total above-ground dry matter （WAGT）, leaf area index （LAI）, leaf N content （NFLV） and weight of seed （WSO）. The degree of variation in model outputs of LAI and WSO were more than 20% and 10% respectively. Among 17 analyzed inputs of ORYZA2000, the model variable of sowing time （EMD） had the highest index of sensitivity on model output. Errors in daily minimum temperature （TMIN）, daily maximum temperature （TMAX） and daily sunshine hour （DHOUR）, i.e. the driving variables of ORYZA200, had much influence on rice yield at mature. The fraction dry matter partitioned to leaves （FLVTB） had much effect on model outputs related to leaf and grain weight, so precision of FLVTB data should be put more attention for reducing uncertainty of yield estimation. The effects of integrating variables with 20% stochastic errors estimated from remote sensed on model outputs （WAGT and WSO） of ORYZA2000 were studied via global sensitivity analysis. Three scenarios of integrating variables （i.e. only LAI, only NFLV, or both of them） were simulated. Among three integrating scenarios, both LAI and NFLV simultaneously integrating with ORYZA2000 showed the highest adjusting effect on simulated WAGT and WSO, LAI alone showed the second highest, and NFLV alone showed the lowest. When WSO and WAGT are estimated integrating ORYZA2000 with variables inverted from remote sensing data, for all integrating scenarios remote sensing data on 70-80th day around after transplanting are more significant that need to be attained, and the remote sensing data before and after this time are also important and should be attained as well. Remote sensing data used for integrate with ORYZA2000 on the period of recovering after transplanting and the later mature of rice have no significance for WSO and WAGT estimation.